Writing
Notes on building AI systems.
Technical writing drawn from real work — retrieval and evaluation, agent reliability, and the statistical discipline behind trustworthy systems. RSS.
-
Purged walk-forward vs ordinary cross-validation: preventing leakage in financial time-series ML
Why k-fold cross-validation leaks on overlapping-label time series, and the purging, embargo, CPCV, and deflated-Sharpe discipline that make evaluation honest.
-
Chunking Arabic legal text for RAG (and why BM25 collapses on clitics)
Why Arabic legal retrieval is hard, how naive BM25 silently fails on clitic-heavy queries, and the morphology-aware chunking and hybrid retrieval that fix it.